Competition and Firm Service Reliability Decisions: A Study of the Airline Industry

Abstract

To understand the impact of competition on organizational service reliability decisions, this study investigates whether firms in the airline industry consider competitors’ actions when making their service reliability decisions. Using data from the U.S. Bureau of Transportation Statistics on flight cancellation rates and average length of flight delays, the authors use two complementary approaches, a simultaneous equation model and a discrete game framework, to examine competitive influence on firm decisions on the level of service reliability. The authors find that competitive effects are asymmetric and differ by the type of firm and its competitors—full-service versus low-cost airlines—as well by level of market concentration. The authors show that internal initiatives, such as on-time bonuses, can substantially improve service reliability but require the firm to account for competitive reactions. Ignoring competitive effects leads to an overestimation of the impact of these programs on service reliability levels.

Keywords

airline industry cancellation rates competition flight delays service reliability simultaneous discrete game simultaneous equation model

After ranking first in on-time arrival among U.S. domestic airlines in 2018, United Airlines touted at its departure gates: “We’re going places (and we’re going on time).”¹ In the airline industry, as in many other services, service reliability, defined as an organization’s “ability to perform the promised service dependably and accurately” (Parasuraman et al. 1988, p. 23) and frequently conceptualized as timely service performance (Cunningham et al. 2004), is “the center pole of service quality” (Berry 2016, p. 4), as it is the most important service dimension in meeting customer expectations. Supporting this claim, Mittal, Ross, and Baldasare (1998) find that negative service experiences that result from low reliability critically relate to customer satisfaction and retention, while Fisher, Gallino, and Xu (2019) find that reduction in delivery time positively influenced online and offline sales in an omnichannel retailing context. Long wait times or unexpected delays lead to customer uncertainty and anger, decreased satisfaction and service evaluations, and negative perceptions of the company image (Sajtos, Brodie, and Whittome 2010; Shirley 1994). Realizing the potential of service reliability to create competitive advantages, many firms allocate resources to programs that improve or motivate service reliability (Li and Lee 1994; Shang and Liu 2011). For example, Black Angus restaurants offer free lunches if meal serving time exceeds ten minutes (Allon and Federgruen 2007), and Bank of America continuously updates its technological platforms and service centers to provide constant and seamless customer access to financial services (Bank of America Corporation 2011).

In the airline sector, service reliability is a frequent theme, highlighted by both government reports and mass media (McCartney 2018; U.S. Department of Transportation 2014), and on-time performance and cancellation rates are important metrics (Allon and Federgruen 2007; Cao et al. 2017) that consumers and investors consider when assessing airline performance (Anderson et al. 2008; Ramdas et al. 2013). Consequently, airlines make tangible investments in tools, such as cushion time adjustment, on-time incentive programs, scheduling optimization, and plane maintenance, to improve these metrics in some or all markets (McCartney 2017; Skobelev 2011). For example, in its Ops Olympics program, American Airlines announced monthly incentives for employees of up to $150 if its service reliability, measured partially by on-time arrival rates, surpassed that of its three largest competitors (Delta Airlines, Southwest Airlines, and United Airlines; Maxon 2014). To achieve better on-time performance records than Delta and United at the Los Angeles Airport (LAX), American Airlines made extensive scheduling adjustments to flights to and from LAX in 2016 (Schlappig 2016).

As the previous examples illustrate, assessing service reliability relative to competitors is prevalent. The performance of competitors and the level of resources of a firm may vary across markets (Mittal, Kamakura, and Govind 2004), yet prior marketing literature on service reliability mostly focuses on understanding the consequences of service reliability devoid of competition (e.g., Mittal, Ross, and Baldasare 1998; Sriram, Chintagunta, and Manchanda 2015). The few studies that do look into the influence of competition on service reliability of airlines (e.g., Mazzeo 2003; Prince and Simon 2015) focus on aggregate competition measures rather than competitors’ market-level actions.

Recognizing the importance of market-level competition in many service industries, such as airlines, banking, hotels, and restaurants, we build on analytical studies on service reliability competition (e.g., Shang and Liu 2011) to empirically examine related decisions in a competitive environment. We suggest that the costs of improving reliability, the impact of service reliability on demand, and predictions of how competitors will respond influence service reliability decisions. Therefore, we assess the impact of competitors’ service reliability on the firm’s service reliability decisions in the airline industry across markets, and thus across different competitive scenarios, while accounting for firm-specific characteristics and market features.

Essentially, we address three interrelated research questions:

In what way, by imitating or differentiating, do firms respond to observed service reliability levels of competitors?

Do response strategies vary with the firm’s own type (i.e., full-service vs. low-cost airlines) and with competitor type?

How do firms’ response strategies change as (a) potential competition changes (i.e., market concentration changes) and/or (b) realized competition changes (i.e., competitive actions change), such that one firm commits to high service reliability and one firm implements a long-term action to improve service reliability?

To address these questions, we propose two complementary approaches. First, using a simultaneous equation model, we evaluate whether competitors’ delays and cancellations affect the service reliability of a focal firm. This is a computationally simple method that allows for interaction between pairs of specific competitors and accounts for the endogeneity of these interactions. Second, we use a discrete game framework (e.g., Bajari et al. 2005; Ellickson and Misra 2008) to investigate competitive effects as a system of simultaneous choices of service reliability while continuing to account for the endogeneity of competitive interactions. This model is more computationally demanding, prompting us to elide the identity of competitors and instead classify the focal firm and competitors into full-service and low-cost airlines; however, it also enables us to evaluate the effects on the market’s service reliability when one firm decides to improve its service reliability.

We apply these two approaches to the U.S. airline industry and use data about two important dimensions of service reliability: flight cancellation rates and average length of flight delays (e.g., Anderson et al. 2008; Cao et al. 2017). From our analysis, we find that (1) for full-service airlines, cancellation and delay decisions of American and Delta tend to follow imitation patterns within each dimension, evidence of intense competition between them; (2) full-service airlines tend to use the cross-service reliability dimensions to differentiate from other full-service competitors: if one firm raises delays, the other reduces cancellations, and vice versa; and (3) the low-cost firm, Southwest, responds to its full-service competitors’ decisions to a lower degree, although its full-service competitors substantially respond to Southwest decisions, mostly by differentiating across service reliability dimensions. These results suggest that competitive responses are asymmetric and strongly depend on the type of the focal firm and competitors.

Given these results and to obtain additional managerial insights, we (1) examine the outcomes of a commitment to a certain service reliability level and (2) measure the impact of particular service reliability programs, similar to American Airlines’ Ops Olympics. A commitment to high service reliability by a full-service firm leads to an increased likelihood (by about 10%) of a low-cost competitor to reduce delay, but this effect reduces in magnitude in markets with more competitors. Finally, we find that one overestimates the direct impact of a service reliability improvement initiative, such as an on-time bonus, for a full-service airline if competitive interactions are not considered. These programs directly affect the service reliability of the focal firm but indirectly affect competitor service reliability efforts. Overlooking these competitive reactions leads to inaccurate evaluations of the profitability of such programs.

Literature Review

Our research relates to three literature streams pertaining to service reliability, the airline industry, and static discrete games. We discuss these each in turn next.

The Importance of Service Reliability

The operations literature has extensively developed analytical models on service reliability. In two seminal papers, Luski (1976) and Levhari and Luski (1978) model duopoly firms’ optimal pricing decisions, given fixed service capacity levels (e.g., number of service counters and service personnel). In related work, Kalai et al. (1992) discuss equilibrium conditions when duopoly firms choose service capacity levels to compete on service speed, given exogenous prices, and De Vany and Saving (1983) model both price and service rate as decision variables and discuss equilibrium conditions in a simultaneous game. These studies establish the existence of equilibria prices and service capacity levels, given exogenous factors such as operational costs or customers’ costs of waiting. Subsequent efforts have modeled the demand side, in addition to the supply side (e.g., Allon and Federgruen 2007; So 2000). To represent demand, So (2000) assumes that customers purchase a service product that maximizes their utility, where utility depends on price and service delivery time, and firms select the levels of price and service delivery time to maximize their profits. The author finds that firms tend to differentiate their service delivery time, conditional on their cost levels, to adjust their service capacity.

Researchers also examine firm optimal decisions about delivery time guarantees, together with decisions on price, service capacity, or service rate using analytical models. For example, So and Song (1998) study a monopoly firm’s joint optimal selection of pricing, delivery time guarantees, and service capacity adjustments when demand is sensitive to both price and delivery time. They recommend longer delivery guarantees when customers emphasize service reliability (i.e., the need for firms to fulfill their guaranteed delivery time), when service capacity decreases, and/or when the cost of increasing service speed increases. Ho and Zheng (2004) prove the existence of Nash equilibria when two firms compete on delivery time guarantees, and they suggest that the choice of delivery-time guarantee depends on the level of service capacity and customer’ sensitivities to guaranteed and actual delivery times. In a related study, Shang and Liu (2011) consider how firms compete by choosing delivery time guarantee levels and on-time delivery rates in industries in which these two factors drive customers’ consumption decisions. If the service capacity is given, the equilibrium condition is service quality differentiation. In a setting in which customers are allowed to switch to competing firms after joining a queue, Li et al. (2012) investigate equilibrium price and service levels and suggest that if the two competing firms use different cost levels to adjust their service capacity, they tend to differentiate their service and price levels.

Empirical research on service reliability is sparse and mainly focuses on understanding the impact of service reliability on demand. Fisher, Gallino, and Xu (2019) use a natural experiment and a difference-in-difference design to provide causal evidence for reduction in delivery time increasing both online and offline sales for an omnichannel retailer, where these beneficial effects seem to exist in the short and long term. Anderson and Mittal (2000) suggest that it is important for firms to understand the negative asymmetry (i.e., the impact of negative performance is greater than that of an equal amount of positive performance) between service performance and customer satisfaction when investing in services. These findings demonstrate the importance of service reliability on retaining customers. Gijsenberg, Van Heerde, and Verhoef (2015) find that performance losses loom larger than gains in the short term and have permanent negative impacts on perceived service quality. Furthermore, Sriram, Chintagunta, and Manchanda (2015) show that variability of service quality increases service termination rates. Png and Reitman (1994) provide an early test of how service time and price at gasoline stations influence demand. They find that, on average, customers are willing to pay about 1% more for a 6% reduction in waiting time. A few articles (e.g., Durrande-Moreau and Usunier 1999; Katz et al. 1991; Lu et al. 2013; Shirley 1994) show that service waiting time decreases service satisfaction and that service reliability affects purchase decisions. None of these studies considers the impact of competitors on the choice of service reliability levels.

Airline Industry and Service Reliability

In the airline industry, previous work used survey methods to establish that service reliability is one of the most important dimensions that passengers cite when evaluating service quality (e.g., Anderson et al. 2008; Cunningham et al. 2002). Key measures of service reliability, flight cancellations and delays, have significant impact on prices and firm stock market performance (e.g., Hansen et al. 2001; Ramdas et al. 2013) as well as operational costs (Irrgang et al. 1995). Hansen et al. (2001) note that flight cancellations result in substantial losses due to costs related to rescheduling aircraft and offering passenger vouchers, with annual estimates in the range of $1 billion to $4 billion). Forbes (2008, p. 1218) quantifies the impact of flight delays on prices, finding that “prices fall by $1.42 on average for each additional minute of flight delay,” with an even greater price response in competitive markets. In studying how service failures due to firm problems may affect customer service evaluations differently from those due to weather conditions, Anderson et al. (2009) find that employee–customer interactions improve customer satisfaction during weather delays but reduce satisfaction when the flight delay is due to firm internal problems. In turn, Ramdas et al. (2013) study the impact of flight delays, cancellations, and other objective service quality indicators on stock returns and find that only long delays have significant negative effects.

Findings on the factors that influence flight cancellations and delays remain inconsistent in the literature. For example, Rupp et al. (2006) find that competitive routes have worse on-time performance, whereas Mazzeo (2003) finds that flight delays are worse in monopoly routes and that competition is associated with better on-time performance. Rupp and Holmes (2006) assert that firms are less likely to cancel flights on competitive routes, and Deshpande and Arikan (2012) find that on-time arrival probabilities increase with competition. Cao et al. (2017) show that flight cancellation rates and delays increase with market concentration, yet delay increases at a slower rate and cancellation rates increase at an faster rate. Prince and Simon (2015) find that incumbents increase flight delay for the entry or entry threats of Southwest, but this pattern does not always apply to other new entrants. In addition, they find mixed evidence for the mechanisms of cost reduction and differentiation. This research considers how variation in market concentration (i.e., the number of firms, market entry) influences the levels of cancellations and delays. However, even though two markets may have the same level of market concentration, competitors’ identities and service reliability choices may differ, leading to varying competitive responses of a focal firm. It is therefore important to model such competitive interactions, and we aim to fill this research gap.

Static Discrete Games

The key advantages of using a discrete game with an incomplete information setup to analyze a competitive setting are that (1) it allows the focal agent (airline) to make strategic decisions while accounting for simultaneous decisions of other agents (competing airlines) and (2) it allows for estimation of a realistic (relatively large) number of players and choice options. The context for the use of such estimation techniques include research in pricing and market entry. For example, Ellickson and Misra (2008) assess how supermarkets choose pricing strategies on the basis of competitor actions, demographics, and store/chain characteristics and find that supermarkets tend to match competitors’ pricing strategies. Ellickson and Misra (2012) incorporate postchoice outcome data into the discrete game estimation and find that cost savings drive the coordination of pricing strategy among supermarkets. Zhu and Singh (2009) examine how Walmart, Kmart, and Target make entry and location decisions, given beliefs about competitor decisions and market and firm characteristics. They find that the three chains appear to have asymmetric impacts on others’ decisions. Vitorino’s (2012) model of strategic entry and location decisions by shopping centers offers strong support for “the agglomeration and clustering theories that predict firms may have incentives to collocate despite potential business-stealing effects” (p. 175). Finally, Orhun (2013) estimates location decisions by two types of supermarkets and finds that competitive pressures from similar supermarkets are much stronger than those from dissimilar types of supermarkets (for a summary of literature, see Table WA1 in the Web Appendix). In this research, we model the service reliability choice of airlines in a market by quarter as a simultaneous-move game, in which each airline has private information and chooses its service reliability level based on market characteristics, firm characteristics in that market, long-term strategic changes, and its beliefs of competitors’ service reliability levels.

Empirical Background and Data

Cancellations and Delays

Competition in the airline industry involves multiple dimensions, such as service reliability, price, comfort and space, and flight availability. The operational resources for airlines include gates and runway slots, aircraft, and crews. The substantial costs of adding aircraft and crews and the constraints on gate and runway availability lead airlines to seek ways to maximize their aircraft and crew utilization by implementing complex algorithms (e.g., Ernst et al. 2004; Lohatepanont and Barnhart 2004). The routing plans suggested by these algorithms tend to work well in ideal conditions, with no interruptions on the runways, gate congestion, crew sickness, strikes, or aircraft wear-out (e.g., McCartney 2014; Saporito 2014), but in reality the ideal is rarely maintained. According to statistics from the U.S. Bureau of Transportation Statistics (BTS), approximately 60% of flight cancellations and delays are due to controllable reasons, such as mechanical problems, shortages of gate slots, crew work time limitations, and unavailability of aircraft.

The reasons for these reliability problems relate closely to the trade-off that firms constantly face: improving service reliability versus reducing the costs of their operational resources (Mittal et al. 2005). According to practitioners (Ewalt 2014; McCartney 2013), the two most important measures of service reliability are the rate of flight cancellations and the average delay time. The BTS collects service reliability data, which include departure delays (difference in minutes between scheduled and actual departure times), arrival delays (difference in minutes between the scheduled and actual arrival times), and cancellations (whether flight is canceled). Since June 2003, the BTS also registers the reasons for delays and cancellations, classified as under the firm’s control or not. For example, carrier and late aircraft–driven delays and cancellations are under the firm’s control, whereas weather, national air system, and security-motivated delays and cancellations are not. These data are available for each domestic flight operated by airlines that earns at least 1% of the revenue market share in the industry. Consistent with previous literature (e.g., Mazzeo 2003; Rupp et al. 2006), we use arrival delays and cancellation rates, attributed to firm actions, to measure service reliability.

Service Reliability as a Continuous Variable or Discrete Choice

Industry managers and practitioners often frame the service reliability challenge as a race among airlines. Some airlines have high standards and good outcome measures; others perform poorly. In 2013, United Airlines marketed its top ranking concerning on-time performance with signs at its gates in airports. Industry magazines and mass media also publish measures of service reliability (Maynard 2013; McCartney 2014) and discuss the few best airlines and the worst airlines based on their recent service reliability, to better inform potential consumers and help them make service choices (see, e.g., Elliott 2018; Ocampo 2019; Park, Vito, and Allen 2019).

Both the percentage of canceled flights and average delays (in minutes) are continuous measures, though most corporate annual reports summarize the firm’s performance as “high-quality” or “average” service, relative to an industry average. For example, Delta noted that it was “at the top of the industry in operational reliability and customer service for major network carriers” (Delta Airlines Inc. 2013, p. 9), Alaska Airlines stated that it “will strive to keep improving productivity while retaining [its] high-quality customer service” (Alaska Air Group 2012, p. 46), and Air Berlin claimed a unique position, offering “above-average service, but with low prices” (Air Berlin Group 2008, p. 43).

With this in mind, we use two approaches to investigate service reliability, reflecting both continuous and discrete aspects. Within a simultaneous equation model, we include delay and cancellation as continuous measures and examine how firms react to each competitor’s service reliability levels. To capture the possibility that firms frame this service reliability competition as a race and choose discrete service reliability levels in response to competitors’ service reliability levels, in a structural discrete game, service reliability is a discrete relative choice: firms choose whether to outperform other airlines serving the same market on different service reliability dimensions.

To create discrete measures, we aim to compare the performance of each airline, in terms of cancellations and delays, with average measures that (1) reflect the market average performance, accounting for regional and seasonal variations, and (2) are not influenced by actions of firms included in the sample. We thus compute the average cancellation rates and delays in the same region combination,² of the same quarter, in the past three years, using full-service and low-cost airlines, other than the four airlines (i.e., American, Delta, United, and Southwest) included in our sample. We classify each firm’s quarterly cancellation rate and delay levels as either above or below the respective average service reliability measures in a market. Such an approach allows for the possibility that all focal airlines in a market can be classified as being all high, all low, or mixed (some high, some low). For each firm, we classify service reliability according to four levels: high service reliability H, when both delays and cancellations are low (i.e., lower than the market average) in a market-quarter; low service reliability L, when both delay and cancellation levels are higher than the market average; and two medium or intermediate levels, such that a firm focuses on having low cancellations but with high delay levels M_c, or low delays but with high cancellations M_d.³ In Table 1, we display the frequencies of the service reliability levels of each firm over all markets and time periods.⁴

Table 1.

Service Reliability Level Frequency by Firm.

	American Airlines	Delta Airlines	United Airlines	Southwest Airlines
High cancellation	3,264	5,289	3,797	15,146
Low cancellation	3,282	9,666	3,343	12,656
High delay	2,465	3,954	2,662	5,787
Low delay	4,081	11,001	4,478	22,015
High cancellation, high delay	1,454	1,870	1,724	3,803
High cancellation, low delay	1,810	3,419	2,073	11,343
Low cancellation, high delay	1,011	2,084	938	1,984
Low cancellation, low delay	2,271	7,582	2,405	10,672

Data Sources and Sample

The T100 Origin and Destination database from the BTS describes traffic for all domestic origins and destinations by firms with annual revenues greater than $20 million and includes variables such as the total passengers served, number of seats, number of departures, and distance for each market, where a market is a unique origin airport–destination airport combination (e.g., Boston Logan International Airport–Seattle-Tacoma International Airport). Data on market characteristics (e.g., number of businesses, income per capita) are available from the U.S. Census Bureau.

We focus our analysis on the larger airlines in the market. We select the three full-service airlines (American Airlines, Delta Airlines, and United Airlines) that constantly had at least 10% of market share in the period of analysis (2003–2018), as well as Southwest Airlines, the leading low-cost airline. We select routes that satisfy the following criteria: (1) combined market share greater than 60% across the four main airlines, so that the selected airlines play important roles in these markets; (2) more than 60% of passengers take a direct flight in that route, so that passenger choices are less affected by connecting flights; (3) only one major airport is located in the origin/destination city, to avoid potential airport-level competition; and (4) the distance between the origin and the destination cities is greater than 300 miles, to avoid the impact of other means of transportation (e.g., car, train, bus) on the findings. Thus, our study offers an analysis of the competition on service reliability by larger players in the airline industry for routes in which direct flights dominate as a means of transportation. The data set includes information about 1,433 routes across 125 cities, flown by the four airlines, during 62 quarters (from quarter 3 of 2003 to quarter 4 of 2018), which produces 56,443 observations. We provide descriptive statistics about these measures of service reliability, firm variables, and market variables in Table 2. The average rate of flight cancellations that are attributable to the firm actions is .31%, and the average length of delay attributed to the firm is 6.5 minutes, with substantial dispersion across markets, firms, and time.

Table 2.

Descriptive Statistics and Correlations.

A: Descriptive Statistics
Variables	Mean	SD	Min	Max
Average flight delay (minutes)	6.539	4.705	0	254
Average flight cancellation rate (%)	.309	.637	0	20
Number of businesses (mean of route cities; in 1,000)	66.34	48.07	6.531	429.9
Yearly income per capita (mean of route cities; in 1,000)	46.48	8.050	24.70	79.69
Competitor’s hub in the market	.188	.390	0	1
Hub origin	.254	.435	0	1
Hub destination	.255	.436	0	1
Quarterly number of flight departures (in 1,000)	.313	.252	.001	1.904
Average capacity utilization	.801	.099	.059	1
Taxi-out time of origin airport (minutes)	14.790	3.005	8.159	32.27
Taxi-in time of destination airport (minutes)	6.466	2.005	2.564	17.62
Origin runway change	.006	.078	0	1
Destination runway change	.006	.078	0	1
Number of seats per flight	150.0	20.800	78	352.4
Number of days with zero or one flight (in 1,000)	.021	.034	0	.092
Relative capacity utilization	.022	.091	–.652	.460
B: Correlations
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)
1. Flight delay	1
2. Flight cancellation rate	.099*	1
3. Number of businesses	.053*	.164*	1
4. Income per capita	.121*	.063*	.382*	1
5. Competitor’s hub	.085*	.063*	.331*	.225*	1
6. Hub origin	−.013	.004	.169*	.090*	−.010	1
7. Hub destination	−.081*	.077*	.168*	.085*	−.012	−.126*	1
8. Flight departures	−.028*	.221*	.258*	.053*	−.030*	.166*	.165*	1
9. Capacity utilization	.013	−.188*	.118*	.257*	.043*	.109*	.114*	−.047*	1
10. Taxi out (origin)	.102*	.018*	.275*	.267*	.242*	.514*	.021*	.122*	.199*	1
11. Taxi in (destination)	−.033*	.087*	.306*	.213*	.286*	−.054*	.453*	.137*	.201*	.036*	1
12. Origin runway change	.035*	.003	−.028*	.035*	−.010	.042*	−.009	−.007	.004	−.039*	−.007	1
13. Destination runway change	.026*	−.002	−.025*	.037*	−.010	−.005	.043*	−.007	.004	−.002	−.028*	−.006	1
14. Seats per flight	.006	−.101*	−.115*	−.024*	.042*	.237*	.225*	−.518*	.064*	−.062*	−.088*	−.008	−.007	1
15. Days with zero or one flight	−.000	.007	.195*	.165*	.104*	−.098*	−.106*	.173*	.278*	.292*	.241*	.003	.002	−.076*	1
16. Relative capacity utilization	−.077*	−.169*	.058*	−.006	−.029*	.033*	.104*	−.031*	.677*	.048*	.156*	−.002	−.012	.076*	.156*

*p < .001.

Determinants of Service Reliability

Despite our focus on the impact of competition on service reliability, we acknowledge other factors that may affect a firm’s decisions. For example, cities connected by a route (market) exhibit considerable variation in the number of businesses and income per capita, which may imply consumer heterogeneity in preferences for service reliability. Several firm-specific characteristics in the market could also influence service reliability. For example, some cities are a hub for an airline, and some airlines offer more flight departures from a certain city, which increases the importance of the markets (i.e., routes) connected to that city for that airline.

In Table 2, we provide the statistics about the different potential determinants of service reliability. For market characteristics, we use (1) the number of businesses, computed as the geometric mean over the number of businesses in the origin metropolitan statistical area (MSA) and destination MSA, across all industries⁵; (2) the income per capita, measured as the geometric mean over the income per capita of the origin MSA and destination MSA, in each year to capture the market’s potential; (3) average time of taxi-in to the destination airport and average time of taxi-out from the origin airport; and (4) additional dummy variables that indicate whether the origin or destination airport is a competitor’s hub, whether there is a change in the number of runway in the origin or destination airport, and whether a market is a monopoly of one of the four considered firms. For firm characteristics, we use measures that correlate with costs and revenues of providing different levels of service reliability, such as the number of flight departures, capacity utilization, number of seats per flight, number of days the firm provides zero or one flight, relative capacity utilization (i.e., average capacity utilization of the focal firm’s flights in the focal route minus the average capacity utilization of the focal firm’s flights departing from the same origin airport), and dummy variables that indicate whether the firm’s hub is at the origin or destination city. Finally, we note other firm-specific long-term strategic changes that may affect service reliability, such as an on-time incentive program offered by United Airlines.⁶

Competition in Service Reliability

Schedule disruptions, resulting from factors such as crew unavailability or delay and mechanical problems, leave firms with reduced operational resources (e.g., fewer runway and gate slots) and, as a result, allow a fraction of flights access to these resources (Forbes and Lederman 2010). Firms choose these flights for operational and economic factors, such as connectivity, demand, and customer goodwill, which vary by markets (Clarke 1998). For example, firms may provide preferential access to resources to flights in markets where consumers have many alternatives, to avoid loss of customers due to flight delays (Mazzeo 2003). Furthermore, firms can make long-term, sticky market-level investments to reduce the impact of schedule disruptions on service reliability in some markets, such as hiring additional ground crews to speed up loading and unloading baggage and acquiring more runway slots (Prince and Simon 2015), as well as short-term, reversible market-level investments, such as assigning larger planes to a market and optimizing flight schedules, aircraft routing, and crew assignment in a market to be more resilient to interruptions (Jhunjunwala et al. 2016; Karp 2015). Schedule disruptions often occur in the airline sector, and firms make decisions to reallocate market-level resources frequently (Forbes and Lederman 2010). In addition, frequent market entries and exits lead firms to adjust resource allocation in the same market (Prince and Simon 2015). Prior research provides empirical support that firms adjust resources on service reliability at the market and quarterly level (e.g., Greenfield 2014; Prince and Simon 2015), and we model these decisions accordingly.

Our observation window includes years when some firms implemented changes that may alter their service reliability levels across markets. This variation in firms’ service reliability, occurring in some years but not others, allows us to identify competitive effects in service reliability. We use two representative markets (Orlando–Los Angeles and San Francisco–San Diego) as illustrations of competitive reactions that are present in our data set. Figure 1, Panel A, shows the delays (for reasons attributed to firm actions) in the Orlando–Los Angeles route. We observe that delays at United dropped sharply in 2009. This drop seems to be a direct result of implementation of an on-time bonus program, which we consider in our analysis. Its competitors, American and Delta, imitating the improvement by United, also show a reduction in their delays for the similar time periods. United decided to abandon the on-time bonus in 2010, which led to a progressive increase in delays for their flights in this route. Here again, we see imitation responses by American and Delta. In the San Francisco–San Diego route (Figure 1, Panel B), the level of delays in United’s flights also dropped around 2009. In this case, the main competitor, Southwest, responded by maintaining or increasing its own delay around that time.

Figure 1.

Evolution in average delays in two markets.

To provide additional examples of model-free data patterns, we selected markets where Southwest and United compete and assessed the service reliability levels of each airline, given the service reliability level chosen by the other. In Table 3, we show the frequency of each classification of service reliability: high; medium, focusing on low cancellation; medium, focusing on low delay; and low. The most common frequency pairs indicate that United chooses the same service level as Southwest or a service level better than Southwest on one dimension. For example, when Southwest offers high service reliability, United most frequently adopts high service reliability; when Southwest chooses medium service reliability with low cancellation, United chooses the same service reliability or high service reliability. These patterns provide some evidence of both imitation and differentiation, at least in one of the service reliability dimensions.

Table 3.

Level of Service: Southwest and United.

Southwest/United	High Reliability	Medium Reliability (Low Cancel)	Medium Reliability (Low Delay)	Low Reliability
High service reliability	17.55%	7.23%	6.08%	4.17%
Medium service reliability (low cancel)	12.36%	13.58%	3.32%	10.09%
Medium service reliability (low delay)	4.01%	.62%	2.40%	1.18%
Low service reliability	3.91%	4.90%	1.15%	7.43%

Modeling Service Reliability and Response to Competition

We use two complementary approaches based on different assumptions of firm behavior: (1) firms’ competitive responses depend on the competitor identity and observed continuous service reliability levels of each competitor; (2) firms’ competitive responses depend on their own type (full-service vs. low-cost) and competitor type but not firm identity, and firms make discrete decisions for their service reliability based on expectations of competitors’ simultaneous service reliability decisions.

Service Reliability as Continuous Variables

Model specification

We define a_dfmt as the level of dimension d = {1, 2,…, D} of service reliability for firm f = 1, 2,…, F at time t = 1, 2,…, T (defined as a quarter in our analysis), in market $m = 1, 2, \dots, m_{f}, m_{f} \in M$ , where m_f represents any route on which firm f operates, and M is the set of routes. In line with prior literature, we model service reliability as a function of firm characteristics and demand factors; new to this research, we also predict that service reliability is a function of each competitor’s decisions related to service reliability.⁷ Finally, we account for firm long-term strategic changes and potential correlations between these service reliability measures, due to unobservable components.

Formally, for a given firm, market, and time period, we have the following D systems⁸ of equations:

(\begin{matrix} a_{1 f m t} \\ a_{2 f m t} \\ ⋮ \\ a_{D f m t} \end{matrix}) = (\begin{matrix} s_{1 f m t} & 0 & \dots & 0 \\ 0 & s_{2 f m t} & 0 & 0 \\ ⋮ & 0 & ⋱ & ⋮ \\ 0 & 0 & \dots & s_{D f m t} \end{matrix}) (\begin{matrix} α_{1 f} \\ α_{2 f} \\ ⋮ \\ α_{D f} \end{matrix}) + \sum_{g \neq f} η_{g m t} \times (\begin{matrix} a_{1 g m t} & 0 & \dots & 0 \\ 0 & a_{2 g m t} & 0 & 0 \\ ⋮ & 0 & ⋱ & ⋮ \\ 0 & 0 & \dots & a_{D g m t} \end{matrix}) (\begin{matrix} β_{1 f g} \\ β_{2 f g} \\ ⋮ \\ β_{D f g} \end{matrix}) + (\begin{matrix} w_{1 f m t} \\ w_{2 f m t} \\ ⋮ \\ w_{D f m t} \end{matrix}) + (\begin{matrix} ϵ_{1 f m t} \\ ϵ_{2 f m t} \\ ⋮ \\ ϵ_{D f m t} \end{matrix}) .

In Equation 1, s_dfmt is a vector of variables that can vary across dimension d and that the firm f takes into account when making service reliability decisions. We include firm f’s characteristics such as its number of flight departures, hub presence, and capacity utilization given the market and time period; its long-term strategic changes (e.g., mergers and acquisitions, or bankruptcy), captured using appropriate firm-year specific dummy variables⁹; and market characteristics such as the number of businesses, income per capita, hub presence of competitors, and an indicator variable for competitive market. The variable η_gmt is a dummy variable indicating whether competitor g is present in market m at time t. The vector a_Dgmt accounts for service reliability choices by its competitors g. A positive value for β_dfg indicates an imitation in competitive response of firm f to competitor g, while a negative value would suggest differentiation. The vector w_dfmt includes market (τ_m), year (τ_y), and quarter (τ_q) fixed effects.

We assume the error term ϵ_dfmt to be independent across firms, time, and markets but to have contemporaneous correlation across service reliability dimensions. Empirically, D = 2, and the two dimensions are the flight cancellation rate and average delay in minutes (both for reasons attributable to the firm).

Model estimation and identification

The identification of the vector β requires that Equation 1 does not have omitted variables that drive both service reliability levels of firm f and its competitors. We include year- and quarter-specific fixed effects to capture the influence of demand seasonality on service reliability and market-specific fixed effects to capture the common factors (e.g., airport conditions, market profitability) that may influence service reliability of all firms operating in market m.

As each firm’s service reliability should depend on the service reliability levels of its competitors in the same market and time, the values of vector a_Dgmt in Equation 1 are functions of a_dfmt, and thus are endogenous. We estimate Equation 1 as a simultaneous equation model using three-stage least squares. For a given market m and time t, each firm’s long-term strategic changes enter only the equations of the focal firm, therefore serving as exclusive restrictions in the focal firm’s equations.

Furthermore, for a given focal firm f at time t, the identities of competitors in market m may be different from the identities of competitors in a different market m'. In addition, for a given focal firm f and market m, the identities of competitors at time t may be different from the identities of competitors at a different time t' because of market entries or exits. These variations in identities of competitors across market and over time facilitate the identification of the vector β. For example, assuming Delta has comparable levels of market characteristics and firm characteristics in two markets in the same quarter, and in one market Delta competes with United while in the other market Delta competes with Southwest, the differences in Delta’s delay and cancellation rate in these two markets reflect the differences in the Delta–Southwest competitive responses from the Delta–United competitive responses.

Finally, while the decision to delay a flight might depend on the airport capacity, the decision to cancel a flight is completely under the control of the firm (Xiong and Hansen 2013). Therefore, some factors can only influence cancellation rate or delay, and it is important to account for these unique factors in the model specification. To take advantage of the efficiency of estimating both dimensions of service reliability in a system (Zellner 1962) and to capture the nuanced differences in cancellation and delay decision processes, we include some covariates that influence only cancellation or delay.

More specifically, in the cancellation equation, we include number of seats per flight, number of days firm f provides zero or one flights in market m in quarter t, and the capacity utilization of flights in market m relative to the average capacity utilization of firm f in markets sharing the same origin airport in the same quarter t. These variables are likely to affect the cost of flight cancellation, which influences cancellation decisions (Xiong and Hansen 2013). For example, canceling a flight with many seats would make rebooking costly, and therefore, a firm may be hesitant to cancel flights if it uses many aircraft with large numbers of seats to serve a market in a given quarter. This variable should not affect delays, because the flight scheduling should already factor in the boarding time, given the size of the aircraft. Similarly, canceling the only flight of the day makes rebooking and accommodation of passengers costly, especially for markets where the airline has sparse capacity. The higher the number of days in which a firm provides one or no flights in a market in a quarter, the costlier it is for the firm to cancel flights in this market. The number of days firm f provides zero or one flights in market m in quarter t, however, should not affect delay, because even if the airline offers one flight in some days, shortage of crews and aircraft mechanical issue may still occur. Finally, if a firm has low capacity utilization in market m but has many flights with higher capacity utilization in markets originating from the same airport, it may cancel flights serving market m to allow flights serving other markets operate on schedule. This variable should not affect delay, as it does not relate to factors of delay, such as mechanical issues or shortage of gates.

In the delay equation, we include dummy variables for runway change in the origin and destination airports, the average minutes for taxi-in of all firms (including smaller airlines) flying to the destination airport, and the average minutes for taxi-out of all firms leaving the origin airport in the current and previous quarter. Runway increase or reduction may change airport congestion, and the two taxi variables can proxy for additional time-varying factors that may affect the airport runway congestion (e.g., temporary runway maintenance). Changes in airport runway congestion, if not properly taken into consideration when a firm schedules flights, may affect flight delays but not affect cancellations, because canceled flights would not experience runway congestions.

Service Reliability as Discrete Decisions

The simultaneous equation model assumes that firms react to the observed continuous level of each service reliability dimension of each competitor. To account for the possibility that firms are uncertain about competitors’ service reliability choices and compete on the basis of expectations of competitors’ average discrete decisions, we propose a discrete game with incomplete information between firms that serve the same market, in line with their type (full-service or low-cost), where each firm chooses the service reliability level (combining the two dimensions) that maximizes its profits in each period. This approach makes it tractable to account for a realistic number of firms and choice options (Orhun 2013) and allows one to contrast firms’ competitive responses given different competitive scenarios (e.g., full-service firms’ responses to full-service competitors vs. full-service firms’ responses to low-cost competitors). Furthermore, discretizing service reliability also accounts for the discrete, categorical aspect of service reliability (e.g., being better than market average) by ignoring tiny variations in service reliability.

We start by outlining the main model assumptions before delving into the model formulation. First, this study takes the number of competitors and their identities as exogenously given. This is a reasonable assumption, because airport entry and flight provision are long-term strategic decisions that airlines make much earlier than operational decisions that drive short-term service reliability variations (Jhunjhunwala et al. 2016).

Second, ideally, revenues minus costs should represent firm profits, but unfortunately information on revenues and costs for each service reliability level are not available. Following previous research (e.g., Ellickson and Misra 2008; Mazzeo 2003; Xiong and Hansen 2013), we assume that observed service reliability choices reflect firm profits—that is, firms choose the service reliability levels that result in the highest profits.

Third, the profits earned from different service reliability strategies might depend on firm characteristics, long-term strategic changes, competitor actions, and market characteristics, similar to the variables considered in the previous approach. Unobserved factors, such as mechanical problems with an aircraft or labor union issues, also influence profits, and we assume these unobserved factors to follow a known distribution (e.g., extreme value distribution). Thus, we compute the choice probability of service reliability levels by integrating over the unobserved term.

Finally, we assume that competition takes place in “local” markets, so the decision for a route is informed by competitors’ decisions for the same route. We also assume that a firm does not observe the service reliability choices of competitors when it selects its own service reliability level but instead has expectations about competitors’ choices. If all firms make rational inferences about competitors’ decisions, competitors’ expected choice probabilities will be consistent with their actual choice probabilities, which are functions of the focal firm’s choice probabilities.

Model specification

At each time period $t = 1, 2, \dots T$ , firm f chooses its service reliability level in a market m. In this framework, we combine the two dimensions of service reliability so that firms make a discrete choice of service reliability from among four options: (1) high service reliability, with low cancellations and delays; (2) a medium level with a focus on low delays but high cancellations; (3) a medium level with a focus on low cancellations but high delays; and (4) low service reliability, with high cancellations and delays. Formally, firms choose a level of reliability (action) a from the action set $A = {H, M_{d}, M_{c}, L}$ , where $H \equiv$ high service reliability, $M_{d} \equiv$ medium service reliability with focus on low delays, $M_{c} \equiv$ medium service reliability with focus on low cancellations, and $L \equiv$ low service reliability. In market m with $F_{m t}$ firms in quarter t, the set of possible actions is $A_{m t} = {H, M_{d}, M_{c}, L}^{F_{m t}}$ with a generic element action $a_{m t} = (a_{1 m t}, a_{2 m t}, \dots, a_{f m t}, \dots a_{F_{m t} m t})$ . The vector of competitor actions in the same market-quarter is $a_{- f m t} = (a_{1 m t}, \dots, a_{f - 1, m t}, a_{f + 1, m t}, \dots a_{F_{m t} m t})$ .¹⁰

For a specific market m and time t, a firm’s state vector is denoted $s_{f m t} \in S_{f m t}$ , with the overall state vector given by $s_{m t} = (s_{1 m t}, \dots, s_{F_{m t} m t}) \in \prod_{f = 1}^{F_{m t}} S_{f m t}$ . The state vector s_fmt is known to all firms and observed by the researcher. It includes long-term strategic changes as well as firm and market characteristics, which we assume to be exogenous. In addition, for each firm f, there are unobserved state variables corresponding to the four possible service reliability options, which are private information to each firm. We denote these unobserved state variables $ϵ_{f m t} (a_{f m t})$ , or simply ϵ_fmt, and represent firm-specific shocks to the profits associated with each choice. With the private information assumption, the simultaneous decisions become a game of incomplete or asymmetric information (Harsanyi 1973), with a resulting Bayesian Nash equilibrium (BNE).¹¹ We assume that the idiosyncratic term ϵ_fmt is drawn from a distribution $g (ϵ_{f m t})$ that is known to all agents, including the researcher.

Firm f maximizes its profits by making service reliability decisions in each market m independently, with the profit function taking the following form:

π_{f m t} = П_{f m t} (s_{f m t}, a_{f m t}, a_{- f m t}) + ϵ_{f m t},

where П_fmt is a known, deterministic function of the states s_fmt and the actions a_fmt and $a_{- f m t}$ . Because of the private nature of the term ϵ_fmt, each firm’s decision rule $a_{f m t} = d_{f m t} (s_{f m t}, ϵ_{f m t})$ is a function of the known state vector and its private shock ϵ_fmt, but it does not depend on the private information of other firms. From the perspectives of both the firm and the econometrician, the probability that firm f chooses action a = k, conditional on the state vector, is

Pr (a_{f m t} = k) = \int 1 {d_{f m t} (s_{f m t}, ϵ_{f m t}) = k} g (ϵ_{f m t}) d ϵ_{f m t},

where $1 {d_{f m t} (s_{f m t}, ϵ_{f m t}) = k}$ is an indicator function equal to 1 if firm f chooses action k, and 0 otherwise.

We denote P_mt as the set of choice probabilities for all firms in market m at time t. Because each firm does not observe competitive actions before making its own choices, we instead account for its expectations about rival actions. The expected profit for firm f when it chooses action a is given by

{\tilde{π}}_{f m t} (a_{f m t}, s_{f m t}, ϵ_{f m t}, P_{m t}) = \sum_{a_{- f m t}} П_{f m t} (s_{f m t}, a_{f m t}, a_{- f m t}) P_{- f m t} + ϵ_{f m t},

where $P_{- f m t} = \prod_{j m t \neq f m t} P_{j} (a_{j} | s_{f m t})$ . With the expectation of these profits, the probability that firm f takes action a in market m at time t is

Ψ_{a, f m t} = Pr [{\tilde{π}}_{f m t} (a_{f m t}, s_{f m t}) + ϵ_{f m t} (a_{f m t}) > {\tilde{π}}_{f m t} ({a^{'}}_{f m t}, s_{f m t}) + ϵ_{f m t} ({a^{'}}_{f m t}), \forall {a^{'}}_{f m t} \neq a_{f m t}],

which is the system of equations that defines the BNE of the game. With components ϵ_fmt drawn from a Type I extreme value distribution, this BNE must satisfy a system of logit equations that constitute the best response probability functions. This framework has been used in several economic settings, and its properties are well understood (e.g., Bajari et al. 2010; Einav 2010; Ellickson and Misra 2008). Brouwer’s fixed point theorem ensures the existence of an equilibrium (see McKelvey and Palfrey 1995).

We next specify the functional form of the expected profits of firm f. We assume that the expected profit of firm f, achieved by choosing the service reliability level k, takes the following form:

{\tilde{π}}_{f m t} (a_{f m t}, s_{f m t}, ϵ_{f m t}, P_{m t}) = s ′_{f m t} α_{k} + \sum_{T = L, F} (ρ_{- f_{F} m t}^{H} β_{k 1}^{T} + ρ_{- f_{F} m t}^{M d} β_{k 2}^{T} + ρ_{- f_{F} m t}^{M c} β_{k 3}^{T} + ρ_{- f_{L} m t}^{H} β_{k 4}^{T} + ρ_{f_{L} m t}^{M d} β_{k 5}^{T} + ρ_{- f_{L} m t}^{M c} β_{k 6}^{T}) + ϵ_{f m t},

where s_fmt includes firm, year, and quarter intercepts, as well as market and firm characteristics¹² of firm f that vary across markets and over time. To account for long-term strategic changes that may influence the profits of choosing certain service reliability levels for more than one period, we review the history of each firm, identify changes (e.g., United’s on-time bonus program, American Airline’s bankruptcy), and include appropriate firm-time fixed effects for these changes in s_fmt. The type T of the focal firm is set to L for low-cost firms and F for full-service airlines. The terms $ρ_{- f_{F} m t}^{H}$ , $ρ_{- f_{F} m t}^{M d}$ , and $ρ_{- f_{F} m t}^{M c}$ denote the expected probabilities that full-service competitors choose high service reliability, medium reliability with low delays, or medium reliability with low cancellations, respectively. The terms $ρ_{- f_{L} m t}^{H}$ , $ρ_{- f_{L} m t}^{M d}$ , and $ρ_{- f_{L} m t}^{M c}$ denote the expected probabilities that low-cost competitors choose high service reliability, medium reliability with low delays, or medium reliability with low cancellations, respectively. The coefficient of each expected probability described previously depends on the type of the focal firm. For identification, we normalize the average profit of the low service reliability level L to 0. We decompose ϵ_fmt as $ϵ_{f m t} = μ_{m} + μ_{f m t}$ , where μ_m is common across all firms in the market m¹³ and μ_fmt is the idiosyncratic for firm f in market m and at time t. This error structure allows for correlated unobserved or unmeasured factors (e.g., air space congestion, touristic market) across firms and time in the same market.

We define the parameter set as $Θ = {α, β, σ}$ and let the variable $γ_{f m t} (k)$ take the value of 1 if action k is chosen and 0 otherwise. With this setup, the optimal choice probabilities for firm f, conditional on unobservables ϵ_fmt, are obtained from

Ψ_{f m t} (a_{f m t} = k | a_{f m t}, s_{f m t}, μ_{m}, P_{m t}, Θ) = \frac{e x p ({s^{'}}_{f m t} α_{k} + \sum_{T = L, F} (ρ_{- f_{F} m t}^{H} β_{k 1}^{T} + ρ_{- f_{F} m t}^{M d} β_{k 2}^{T} + ρ_{- f_{F} m t}^{M c} β_{k 3}^{T} + ρ_{- f_{L} m t}^{H} β_{k 4}^{T} + ρ_{f_{L} m t}^{M d} β_{k 5}^{T} + ρ_{- f_{L} m t}^{M c} β_{k 6}^{T}) + μ_{m})}{\sum_{j = {H, M_{d}, M_{c}, L}} e x p ({s^{'}}_{f m t} α_{j} + \sum_{T = L, F} (ρ_{- f_{F} m t}^{H} β_{j 1}^{T} + ρ_{- f_{F} m t}^{M d} β_{j 2}^{T} + ρ_{- f_{F} m t}^{M c} β_{j 3}^{T} + ρ_{- f_{L} m t}^{H} β_{j 4}^{T} + ρ_{f_{L} m t}^{M d} β_{j 5}^{T} + ρ_{- f_{L} m t}^{M c} β_{j 6}^{T}) + μ_{m})}

The likelihood is given by the following expression:

L = \prod_{t \in T} \prod_{m \in M} \prod_{f \in F_{m t}} {[Ψ_{f m t} (a_{f m t} = k | a_{f m t}, s_{f m t}, μ_{m}, P_{m t}, Θ)]}^{γ_{f m t} (k)}

s . t . P_{m t} = Ψ_{m t} (s_{f m t}, P_{m t}, Θ, μ_{m}),

in which $Ψ_{m t} (s_{f m t}, P_{m t}, Θ)$ denotes the systems of choice probabilities of every firm for each possible service reliability level. This likelihood function involves a system of discrete choice equations that satisfies the set of fixed-point constraints ( $P_{m t} = Ψ_{m t}$ ).

Model estimation and identification

We use a two-step conditional choice probability estimator (e.g., Bajari et al. 2005; Ellickson and Misra 2008) to obtain the parameter vector Θ. In a first stage, the estimation procedure obtains a consistent estimate of $P_{m t}$ , the probabilities that are part (implicitly) of Equation 7. Within each dimension of the state space (i.e., a unique combination of state variables; each state variable is categorized into high and low levels), we obtain the beliefs of each firm choosing each service reliability level nonparametrically by computing the observed average choice probabilities in the data. With these estimates of beliefs of $P_{m t}$ , we have the necessary information to obtain Ψ_fmt, which then can be plugged into the likelihood function. We maximize this likelihood in the second stage of the estimation. Then, we use the in-sample predictions from the maximum likelihood estimation stage to update firm beliefs about competitors’ service reliability probabilities and perform the maximum likelihood estimation with these updated beliefs. We proceed iteratively across these two steps until we achieve consistent estimates of Θ. Newey and McFadden (1994) demonstrate consistency and asymptotic normality for two-step estimators in general, and Bajari et al. (2005) offer formal results for the proposed approach. Because of this two-step approach, we cannot use the inverse information matrix for standard errors of the parameters and apply a bootstrap approach instead.

Prior static game literature details the identification strategy for similar models (e.g., Bajari et al. 2010). Three assumptions must be satisfied. First, the private information (μ_fmt) has to be distributed i.i.d. across firms and actions in any market and drawn from a distribution of a known parametric form. Second, the expected profit of one strategy must be normalized to 0. We normalize the expected profits of the lowest level of service reliability L to be 0. Third, an exclusion restriction must be satisfied to identify the deterministic part of the profit function, given choice probabilities and the expected profits. The choice probability of each firm is a function of its beliefs about the conditional probabilities of its competitors as well as state variables. If the variables that enter to define the beliefs are the same as those in $s'_{f m t} α_{k}$ , then there is no exclusive variation of each equation to help identify the effects. To solve it, we ensure that several covariates enter the payoff function of firm f in market m at time t but do not enter the payoff functions of other firms in the same market and time. The private shocks μ_fmt, which are idiosyncratic for firm f in market m at time t, create an exclusion restriction, and, among the included independent variables, we have firm fixed effects and firm-level long-term strategic changes entering solely the respective firm’s profit function. In addition, values of number of flight departures and capacity utilization vary by firm within the same market and time, therefore introducing additional variation to the payoff function of each firm in the same market and time.

We highlight the significant variation in the data that helps identify the impact of competition and other variables. For example, a firm might consistently opt for a low level of service reliability in most markets, but we occasionally find deviations from this strategy, such that it pursues a better service level when its competitors exhibit low or high delays (or cancellations) in the particular market. This variation reveals pressure from competitor decisions, which prompt the firm to go for a less chosen option and shed light on the magnitude of the effects needed to produce this variation. In addition, we include monopoly markets, duopoly markets, and oligopoly markets in the sample, in which the set and number of firms also changes over time and over markets. Finally, although we consider only the types (full-service vs. low-cost) and ignore the identity of competitors in this approach, the variation in service reliability levels due to the presence or absence of competitors will also reflect the strength of competitor effects.

We estimate a discrete game with incomplete information. In this type of games, beliefs of firms are “monotonic, continuous, and strictly bounded between 0 and 1, so existence of a solution to the system of equations follows immediately from Brower’s fixed-point theorem” (Vitorino 2012, p. 185). Grieco (2014) also proves that private information reduces the threat of multiple equilibria. As we use the two-step method for estimation incomplete information games, proposed by Aguirregabiria and Mira (2007), our estimation follows their assumption that the data are generated by only one equilibrium. Thus, despite the possibility of potential multiple equilibria, the equilibrium that has been selected by players or nature will be identified from the conditional choice probabilities in the data (Aguirregabiria and Mira 2007). We then followed Vitorino (2012) to impose a follow-up assumption that players do not switch to other equilibria as long as the primitives of the model (i.e., the parameter vector Θ and the explanatory variables) do not change. Finally, potential for multiple equilibria also decreases with the introduction of asymmetry among players (Grieco 2014; Orhun 2013). Our firm fixed effects and firm-level long-term strategic changes address this point by introducing asymmetry among players.

Results

In Tables 4 –6, we present results for our models. In terms of overall model fit, the R-squares of the simultaneous equations range from 8.3% (cancellation equation of United) and 28.3% (the delay equation of Southwest), and the hit ratio of the discrete game model is 47.8%.

Table 4.

Results for the Simultaneous Equation Model.

A: Firm and Market Characteristics
	Cancellation Rate				Delay
	American	Delta	United	Soutdwest	American	Delta	United	Soutdwest
Intercept	−.008(.012)	−.050***(.014)	−.162***(.015)	.103***(.016)	−.274***(.073)	.107(.086)	−.512***(.106)	−.448** (.143)
Characteristics of the Focal Firm
– Number of flight departures	−.050(.049)	−.068**(.031)	.203***(.043)	.210***(.003)	−1.950***(.250)	−1.560***(.160)	−.742***(.267)	.343 (.228)
– Hub origin	−.150***(.020)	−.032*(.019)	−.129***(.031)	.037(.050)	.670***(.118)	.202*(.109)	2.112***(.214)	.444 (.469)
– Hub destination	.032*(.019)	.045**(.019)	.069**(.031)	.086*(.049)	.130(.113)	−.053(.111)	2.686***(.211)	−.170 (.459)
– Capacity utilization					4.349***(.304)	2.159***(.254)	4.860***(.463)	3.212***(.248)
– Number of seats per flight	.002***(.000)	−.002***(.000)	−.005***(.000)	−.004***(.001)
– Number of days with zero or one flight	.061(.163)	−.303**(.134)	.009(.196)	−.072(.102)
– Relative capacity utilization	−.879***(.052)	−.744***(.037)	−1.401***(.060)	−.760***(.027)
Market Characteristics
– Markets with at least one large competitor	−.001(.010)	−.006(.011)	.062***(.011)	−.061***(.012)	−.128**(.055)	.142**(.062)	.275***(.079)	−.278*** (.107)
– Competitors’ hub	.101***(.012)	−.088***(.015)	.140***(.019)	.062***(.012)	.497***(.074)	.150*(.086)	.094(.130)	.840***(.114)
– Total number of businesses	−.003***(.000)	3.32E−04(5.03E−04)	−.004***(.001)	2.14E−04(5.79E−04)	−.003(.003)	.006**(.003)	.007*(.004)	.025***(.005)
– Income per capita	.002***(.001)	−.002(6.91E−04)	−.005***(.001)	.003**(.001)	−.006(.004)	−.003(.004)	.014**(.005)	.004 (.008)
– Taxi out time of origin airport					−.032***(.009)	.148***(.009)	.036***(.013)	.172***(.018)
– Taxi in time of destination airport					.049***(.010)	−.047***(.012)	.082***(.015)	102***(.020)
– Origin runway change					.024(.093)	−.016(.108)	.071(.133)	.129 (.185)
– Destination runway change					−.006(.093)	−.020(.107)	.029(.132)	.002 (.185)
B: Long-Term Strategic Changes.
	Cancellation Rate				Delay
	American	Delta	United	Southwest	American	Delta	United	Southwest
Long-Term Strategic Changes of the Focal Firm
– Bankruptcy of United			−.482***(.017)				−1.978***(.120)
– Bankruptcy of American	.050***(.014)				−.997***(.086)
– Merger of Delta and Northwest		.071***(.014)				1.026***(.083)
– Merger of American and US Airways	−.243***(.014)				−.250***(.086)
– Merger of United and Continental			−.526***(.016)				−2.478***(.113)
– Merger of Southwest and AirTran				.066***(.014)				.278**(.128)
– Atlanta airport schedule redesign by Delta		.131***(.014)				.721***(.079)
– Service expansion by Delta		−.369***(.009)				−.807***(.055)
– Tech investment by American	.119***(.013)				.889***(.078)
– On-time bonus by United			−.518***(.020)				−3.771***(.139)
– Tech investment by Southwest				−.034**(.015)				1.309***(.132)
– Large fleet replacement by Southwest				.047***(.010)				−.769***(.086)

Notes: This specification includes year, quarter, and market fixed effects.

*p < .1.

**p < .05.

***p < .01.

Table 5.

Results of the Discrete Game.

A: Market and Firm Characteristics and Long-Term Strategic Changes
	High Reliability		Low Cancellations		Low Delays
	Coef.	SE	Coef.	SE	Coef.	SE
Intercept	1.246***	(.200)	1.031***	(.179)	.475**	(.186)
Firm Characteristics
Number of flight departures	−.873***	(.090)	−.716***	(.071)	−.414***	(.088)
Hub origin	−.033	(.087)	.089	(.077)	−.299***	(.088)
Hub destination	−.226***	(.085)	−.297***	(.070)	.112	(.089)
Capacity utilization	.471***	(.044)	.409***	(.040)	.098**	(.048)
Market Characteristics
Markets with at least one large competitor	−.174	(.311)	−.129	(.297)	−.472	(.295)
Full-service focal and full-service competitor(s)	−.827	(.512)	−.028	(.422)	−1.159	(.539)
Full-service focal and low-cost competitor(s)	−2.886***	(.882)	−1.352*	(.765)	−2.405**	(1.057)
Low-cost focal and full-service competitor(s)	−.287	(.559)	.471	(.456)	.402	(.597)
Competitors’ hub	−.560***	(.145)	−.297**	(.128)	−.471***	(.126)
Total number of businesses	−.415***	(.073)	−.339***	(.075)	−.125*	(.073)
Income per capita	−.192***	(.073)	−.282***	(.065)	−.008	(.067)
Origin runway change	−.518**	(.237)	−.396**	(.165)	−.035	(.300)
Destination runway change	−.234	(.220)	−.051	(.196)	−.515	(.352)
Long-Term Strategic Changes
Bankruptcy of United	1.793***	(.181)	.806***	(.187)	1.436***	(.193)
Bankruptcy of American	−.747***	(.251)	−.407	(.249)	−.410	(.259)
Merger of Delta and Northwest	.452**	(.199)	.129	(.195)	.148	(.207)
Merger of American and US Airways	.558**	(.275)	.610**	(.276)	.931***	(.311)
Merger of United and Continental	1.163***	(.343)	1.006***	(.255)	.866***	(.322)
Merger of Southwest and AirTran	−.937***	(.223)	−.107	(.223)	−.651**	(.253)
Atlanta airport schedule redesign by Delta	.124	(.157)	−.070	(.162)	.418**	(.189)
Service expansion by Delta	.822***	(.237)	1.107***	(.237)	.209	(.248)
Tech investment by American	−.712***	(.166)	−.567***	(.169)	−.505**	(.199)
On-time bonus by United	2.352***	(.244)	.921***	(.212)	1.843***	(.276)
Tech investment by Southwest	−1.148***	(.132)	−.727***	(.137)	−.194	(.158)
Large fleet replacement by Southwest	−.142	(.118)	−.410***	(.125)	.293**	(.139)
B: Competitive Strategies
	High Reliability		Low Cancellations		Low Delays
	Coef.	SE	Coef.	SE	Coef.	SE
Response by full-service firms (United, American, Delta) to…
High reliability of full-service competitors	.440	(.439)	.424	(.441)	.789	(.530)
High reliability of low-cost competitor	3.105***	(.653)	1.300**	(.561)	2.841***	(.790)
Low delay of full-service competitors	3.828**	(1.664)	1.199	(1.089)	3.723***	(1.428)
Low delay of low-cost competitor	4.389**	(2.205)	2.621	(1.956)	4.842*	(2.646)
Low cancellations of full-service competitors	2.103**	(.891)	.874	(.653)	2.638***	(.990)
Low cancellations of low-cost competitor	4.327***	(1.070)	1.882**	(.947)	3.574***	(1.326)
Response by a low-cost firm (Southwest) to…
High reliability by full-service competitors	.775	(.596)	−.053	(.502)	1.033**	(.523)
Low delay by full-service competitors	1.852	(1.308)	.370	(1.001)	−.543	(1.378)
Low cancellations by full-service competitors	.613	(.922)	−.528	(.808)	−2.252*	(1.191)

Notes: The baseline is the low service reliability condition, with high delays, high cancellations, which takes the value of zero for identification.

*p < .1.

**p < .05.

***p < .01.

Table 6.

Results of the Simultaneous Equation Model: Competitive Responses.

	Delay Regression				Cancellation Regression
	(1) Cancellation Rate		(2) Delay		(3) Cancellation Rate		(4) Delay
	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
American Airlines response to…
Delta Airlines	−.547***	.085	.280***	.033	.055***	.014	.002	.004
United Airlines	.178*	.099	−.090***	.020	.104***	.016	−.023***	.004
Southwest	.941***	.093	.034*	.018	.078***	.015	.001	.003
Delta Airlines response to…
American Airlines	.638***	.144	.298***	.024	.066***	.025	−.023***	.004
United Airlines	−.064	.115	.051**	.023	.005	.019	.013***	.004
Southwest	.456***	.104	.034*	.019	.119***	.018	.018***	.003
United Airlines response to…
American Airlines	−1.132***	.194	−.083**	.032	−.115***	.028	−.006	.005
Delta Airlines	−.165	.125	−.004	.049	.010	.017	−.015***	.006
Southwest	1.169***	.133	−.105***	.026	−.081***	.019	.003	.003
Southwest response to…
American Airlines	1.759***	.244	.042	.043	.258***	.028	.037***	.005
Delta Airlines	.152	.177	−.233***	.069	.142***	.020	.005	.006
United Airlines	−.222	.206	−.025	.042	−.177***	.023	.050***	.005

Notes: This specification includes year, quarter, and market fixed effects.

*p < .1.

**p < .05.

***p < .01.

Firm Characteristics and Market Variables

We present the results in Panel A of Tables 4 and 5. We find statistically significant and consistent results for firm characteristics across both approaches. Specifically, in a market with a hub airport, whether in the origin or destination city, most airlines tend to exhibit lower service reliability. These results validate and extend the findings of Mayer and Sinai (2003), who indicate that hub airports suffer longer delays than nonhub airports, and hub carriers experience flight delays because they “cluster their flights in short spans of time to provide passengers with a large number of potential connections with a minimum of waiting time” (p. 1194). Results of the simultaneous equation model show that the number of days with zero or one flight reduces cancellation rates for Delta (−.303, p < .05), which is corroborated by the results of the discrete game approach, where a higher number of flight departures reduces the attractiveness of high or medium service reliability.

Capacity utilization leads airlines to prefer high or medium service reliability with low cancellations in the discrete game approach, a finding confirmed by the negative coefficients of relative capacity utilization on the cancellation rate of all firms in the simultaneous equation model. In contrast, we find positive effects of capacity utilization on delays in the simultaneous equation model. This finding seems reasonable: airlines want to avoid canceling better utilized flights, and when it is not feasible to take off on time, they choose to delay as opposed to cancel those flights. Finally, the number of seats per flight negatively affects the cancellation rates of Delta, United, and Southwest, largely in agreement with the findings of Xiong and Hansen (2013).

We included the following market variables to explain service reliability choices: level of competition (i.e., whether the focal firm has at least one large competitor in the market), competitor’s hub, the number of businesses, income per capita, and runway changes in the origin or the destination airport. Again, in both approaches, the results are largely consistent. Consistent with extant literature (Mayer and Sinai 2003; Xiong and Hansen 2013), if a city is a competitor’s hub, we find that most airlines have higher cancellation rates and/or higher delays. In addition, consistent with Pai (2010), results of the discrete game show that as the income of customers and number of businesses in the market served increase, profits and service reliability levels diminish.

Long-Term Strategic Changes

We study the impact of 12 firm-specific long-term strategic changes on service reliability choices. We find that the on-time bonus of United and the service expansion of Delta have clear positive implications for service reliability in both models. In the simultaneous equation model, their coefficients (Table 4, Panel B) for both delays and cancellation rates are negative. In the discrete game results (Table 5, Panel A), United’s on-time bonus made the high service reliability (2.352, p < .01), and medium service reliability with low cancellations (.921, p < .01) or low delays (1.843 p < .01) preferable for the company. The service expansion led Delta to prefer high service reliability (.822, p < .01) and medium service reliability with low cancellations (1.107, p < .05) rather than low service reliability.

The merger of American and US Airways increased the likelihood that the merged American would adopt high service reliability (.558, p < .05) or medium service reliability with low cancellations (.61, p < .05) or low delays (.931, p < .01) rather than low service reliability. Similarly, the merger of United and Continental increased the likelihood of merged United to adopt high service reliability (1.163, p < .01) or medium service reliability with low cancellations (1.006, p < .01) or low delays (.866, p < .01). In addition, United became more likely to adopt high service reliability (1.793, p < .01) or medium service reliability with low cancellations (.806, p < .01) or low delays (1.436, p < .01) after its bankruptcy. Finally, its large fleet replacement led to increased profits for Southwest when it engaged in medium service reliability with low delays (.293, p < .05) and decreased profits for Southwest when it engaged in medium service reliability with low cancellations (−.41, p < .01), probably due to the substantial transition involved with this decision, which is also reflected in its estimates in the simultaneous equation model (.047, p < .01; −.769, p < .01).

The Impact of Competitors’ Decisions on Service Reliability

In Table 6, we provide the estimates of competitive impact using the simultaneous equation model, while Table 5, Panel B, presents the results for the discrete game approach. Given that the dependent variables in Table 6 have different scales (minutes for delays and percentages for cancellation), we display in Figure 2, Panels A–D, the resulting elasticities for the competitors’ effects for both dimensions of service reliability. We use the results in Table 5, Panel B, and Figure 2 to draw conclusions on the impact of competitors’ decisions on service reliability.

Figure 2.

Competitor own-elasticities and cross-elasticities of delay and cancellation rates.

Starting first with Southwest, the only low-cost airline in our analysis, we find that this firm’s delay responds the least to the decision of the three full-service airlines. In Figure 2, its delay elasticities are in the far-right columns in Panels A and D. In most cases, the elasticities have the lowest magnitude when compared with other firms. Southwest’s cancellation responds to the decision of full-service airlines mostly by differentiating, by increasing the level of cancellations either when a full-service competitor decreases it or when a full-service competitor increases delay.¹⁴ In terms of the discrete game, we observe in Panel B of Table 5 close to no significant response by Southwest to the decisions of full-service airlines, except if the full-service competitors offer high service reliability—then Southwest leans toward providing medium service reliability with low delay. These results seem to indicate that Southwest may consider itself a differentiated product from the full-service airlines and thus not competing directly with them (or competing to a lower extent), even though all are large players in the market.

Among the full-service airlines, we observe interesting elasticity patterns in Figure 2. Within each dimension, we mostly observe an imitation pattern of American and Delta. This is clear in Figure 2, Panels A and B, where almost all elasticities of these two firms of the simultaneous equation model are positive. United, in contrast, tends to prefer to differentiate from American and Delta, as suggested by its negative elasticities. The results of the discrete game are consistent with those of simultaneous equation model, as the imitation pattern of American and Delta is reflected in the positive coefficients of full-service competitors’ medium service reliability with low delays on a full-service focal firm’s profits of providing high service reliability (3.828, p < .05) and medium service reliability with low delay (3.723, p < .01), compared with the baseline choice (coefficient equal to zero) of offering low service quality.

Across dimensions, however, firms show mostly a differentiating pattern, displayed in Figure 2, Panels C and D: if one firm increases delays, competitors respond by decreasing cancellation rates, and vice versa. The discrete game provides similar results to simultaneous equation model, with firms responding to a focus by a competitor on one-dimension of service quality by increasing service reliability to the highest level, or by responding to low cancellations with low delays. Interestingly, when a full-service company does achieve a high service reliability, the other full-service competitors do not have a significant dominant strategy, possibly indicating that they focus on other dimensions besides service reliability as competitive tools. Finally, full-service firms respond to Southwest mostly by differentiating across dimensions (i.e., decreasing cancellation rates as a response to a decrease in delays).¹⁵

Overall, we summarize the findings of the competition effects across service reliability dimensions in the following four statements:

Among full-service airlines, the cancellation and delay decisions of American and Delta tend to follow imitation patterns, within each dimension;

Full-service airlines tend to use the cross-service reliability dimensions to differentiate from other full-service competitors: if one firm raises delays, the other reduces cancellations, and vice versa;

The low-cost firm, Southwest, responds to its full-service competitors’ decisions to a less degree, although its full-service competitors significantly respond to Southwest decisions, mostly by differentiating across service reliability dimensions.

These results suggest that competitive responses are asymmetric and depend on the type of the focal firm and competitors.

Competitive Responses and Market Concentration

Inspired by findings in prior research that airline on-time performance varies by market concentration (e.g., Cao et al. 2017; Mazzeo 2003), we explore whether competitive responses differ in high-concentration markets from low-concentration markets. We follow Brueckner (2002) and define a market to be high concentration if the flight share of the largest carrier exceeds 65% and define a market to be low concentration if otherwise. We then split the sample based on market concentration and estimate the two models using these two subsamples. We present the results in Tables WA7–WA11 in the Web Appendix.

Comparing the coefficients of competitor’s delays on the focal firm’s service reliability from the simultaneous equation model (see Table WA7), we find that the competition strategies between full-service airlines and Southwest change as market concentration varies. For example, in low-concentration markets, Southwest’s delay significantly affects both delays and cancellation rates of American, Delta, and United. Furthermore, Southwest’s delay has positive effects on the delay of all three full-service airlines. Conversely, in high-concentration markets, Southwest’s delay has significant and positive effects only on the service reliability of Delta. These results suggest that full-service airlines respond to Southwest more in low-concentration than in high-concentration markets. The discrete game estimation suggests similar differences in service reliability competition (see Web Appendix Table WA11): more coefficients of competitor strategies (especially those of low-cost competitors) are significant in low-concentration markets than in high-concentration markets, and the low-concentration markets exhibit more imitation patterns than high-concentration markets.

Managerial Insights

We highlight our contribution with two additional managerial issues. We evaluate (1) the change on the overall market service reliability level when one airline commits to a particular service reliability level and (2) the effect of service reliability programs on the focal firm and its competitors.

Commitment to a service reliability level

This analysis is inspired by the fact that starting in 2011, Delta Airlines actively strove to improve its service reliability, using new analytical software and instruments to monitor the health of aircraft and facilitate interdepartmental communications to speed up flight rescheduling after unexpected disturbances. We use the discrete game approach to measure the impact of this commitment to high service reliability. To do so, we fix Delta Airlines’ service reliability to a high level, assume that Delta does not react to competitors, and assume that competitors observe this choice, such that their expectations about Delta become certain for this service level. To illustrate the impact of this decision, we select two distinct markets (Atlanta–Boston and Minneapolis–Denver) and analyze the service reliability choice probabilities.¹⁶ We note that even though Southwest reacts less to the full-service competitors’ actions, the exception previously mentioned is when a competitor reaches a high level of service quality, which this analysis exemplifies. Thus, we expect a significant reaction from Southwest.

In the Atlanta–Boston market, Delta has one major competitor, Southwest. When Delta commits to high service reliability, Southwest’s choice probability associated with providing medium service with low delays increases. That is, if Delta commits to high service reliability, the average likelihood that Southwest chooses a medium service level with low delays increases from 11.3% to 21.2%, at the expense of other three service reliability levels, each of which decreases by 3%–4%.

In the Minneapolis–Denver market, Delta, Southwest, and United compete. If Delta keeps its commitment to high service reliability, Southwest focuses on increasing its medium service reliability with low delays, largely at the expense of reducing its choice of high service reliability, while United focuses on increasing its medium service reliability with low cancellations and low service reliability at the expense of the other two service reliability levels. Specifically, the average choice probability for medium service reliability with low delays by Southwest increases from 8.4% to 12.1%, and the average choice probability for high service reliability of Southwest decreases from 27.3% to 25.3%. For United, the average choice probability for medium service reliability with low cancellation rate (low service reliability) increases from 30.2% (33.2%) to 34.2% (39.5%), while the average choice probability for medium service reliability with low delay (high service reliability) decreases from 13.1% (23.5%) to 9.6% (16.6%). This result indicates a similar differentiation strategy of Southwest, even though the competitive reactions are weaker in markets with more competitors. This weaker response reflects the interactions between Southwest and United: given the service reliability choice of Delta, Southwest increases the probability of medium service reliability with low delays at the expense of high service reliability. These changes by Southwest make United less inclined to provide high or medium service reliability with low delays, which in turn reduces the likelihood that Southwest will choose medium service reliability with low delay.

In conclusion, a commitment to improving service reliability of both dimensions by Delta motivates the low-cost competitor to focus on the delay dimension of service reliability at the expense of the cancellation rate. The extent of the reaction depends on the number of competitors.

Evaluating service reliability improvement strategies

We quantify the impact of the on-time bonus program implemented by United Airlines during 2009 and 2010 by comparing the actual situation with a counterfactual scenario with no on-time bonus program. We then compare the latent profits earned by United and competing airlines based on their service reliability choices. To create the counterfactual setting, we changed the dummy variable that captures the on-time bonus at United from 1 to 0 for the duration of the program and compute the resulting choice profits and probabilities for both United and its competitors. For this illustration, we consider the Denver–San Francisco and Denver–Los Angeles markets.

In the Denver–San Francisco market, United and Southwest compete, and the on-time bonus makes it more profitable for United to provide high or medium service reliability levels. In the actual scenario with on-time bonus, compared with the counterfactual scenario without it, United is more likely to provide high or medium service reliability with low delays. Latent profits of high service reliability, relative to low service reliability, increase from −.43 to 2.03, and the respective choice probability increases from 41.6% to 43.2%, while latent profits of the medium service reliability with low delays increase from −.50 to 2.01, and its choice probability increases from 32.8% to 42%.

The higher likelihood by which United provides high or medium service reliability with low delays with the on-time bonus, compared with the scenario without it, makes it more profitable for Southwest to pursue high service reliability: the latent profits of high service reliability increase from 1.16 to 2.98, and its choice probability increases significantly. Therefore, the on-time bonus not only improves the service reliability of United Airlines but also motivates its competitor, Southwest, to improve in this market.

In the Denver–Los Angeles market, the model predicts a similar result, and the program has significant positive results for United Airlines. In the scenario with the on-time bonus compared with one without it, United’s high service reliability probability increases to 38.1% from 44.8%, and its medium service reliability with low delays probability increases to 31.1% from 52.4%. In terms of competitor effects, Southwest would react by a higher probability of offering high service reliability by 33%. The other competitor, American Airlines would increase the probability of offering high service reliability and medium service reliability with low delays by 28.3% and 21.8%, respectively.

Thus, the implementation of the on-time bonus leads to improvements in United Airlines’ service reliability, and competitors are more likely to provide service reliability levels with low delays. Given this competitive response, United Airlines further improves its likelihood to provide high or medium service reliability with low delays. Put differently, the service reliability improvements of United emanate from (1) a direct impact of the implementation of the on-time bonus program and (2) the indirect effect of competitive responses.

A model that does not consider competitor reactions would ignore the second process and attribute all of United’s service reliability improvements to the implementation of the on-time bonus program alone. We establish that the direct impact of an on-time bonus initiative would be overestimated if one would ignore competitive interactions. To measure this, we reestimated the discrete game model without competitive effects and find that the coefficient of the on-time bonus program is higher by 4% for high service reliability and for median service reliability with low delays, when compared with our model with competitive effects. An approach that does not account for competitive interactions misattributes the improvement of service reliability to internal factors, instead of correctly assessing the competitive response.

Conclusion

Our primary contribution is to the service reliability literature. Prior empirical studies related to service reliability focus on the demand side, testing the costs or consequences of customer waiting (e.g., Allon et al. 2011; Shirley 1994). We instead examine the supply side and aim to explain firms’ competitive decisions. Some previous analytical studies note the equilibrium conditions that arise when two firms compete on service reliability and suggest that, given different capacity costs, firms tend to differentiate their service levels (e.g., Li et al. 2012; Shang and Liu 2011). Our research methods and context enable us to advance empirical understanding of service reliability competition by teasing out the effects of firm characteristics (e.g., capacity utilization) on service reliability choices and focusing on the effects of competitors’ service reliability choices on the choices of a focal firm. We find evidence of substantial competitive reactions to full-service and low-cost competitors and quantify the impact of service reliability improvements considering competitors’ responses.

Our research also enriches literature on the airline industry (e.g., Deshpande and Arikan 2012; Mayer and Sinai 2003; Rupp and Holmes 2006). Extant literature offers mixed findings about service reliability, suggesting that competition intensity either decreases delays and cancellations (e.g., Cao et al. 2017; Mazzeo 2003) or increases them (e.g., Prince and Simon 2015; Rupp et al. 2006). However, these studies did not look into the precise competitive interactions among firms, such as how focal firms might choose their flight cancellations and delays on the basis of their competitors’ expected choices. The contradictory findings from prior research may have arisen because firms’ service reliability choices in a market depend on their competitors’ service reliability choices, in addition to competitive intensity. We deepen understanding of how competition influences firms’ service reliability choices by explicitly modeling their competitive interactions. In terms of managerial implications, our results suggest that firms’ competitive reactions vary with the type of firm, competitors, and market concentration.

It is important to recognize some limitations of our research. Due to data limitations, we use flight cancellations and delays to represent service reliability in the airline industry, but these two constructs may not be sufficient to capture the entire service reliability concept. The discretization of outcomes in the discrete game model also limits its conclusions and the significance of the competitive results, which suggests the need for complementary approaches that can exploit the original, continuous nature of the data, as in our simultaneous equation model. Service reliability decisions may have implications for more than one period. We include major firm-level long-term strategic changes (e.g., mergers and acquisitions, crew scheduling redesigns) that may influence the profits of certain service reliability choices for more than one period in both models. However, we do not model dynamic competitive interactions among firms. Our empirical methods do not allow us to separately identify the effectiveness of easy-to-reverse actions and sticky actions on service reliability, despite controlling for long-term strategic changes that may affect service reliability. We encourage further research to address these issues. Finally, because airlines operate in multiple markets, they might make their service reliability investment decisions for a set of markets as opposed to market by market. A firm’s service reliability of a market depends on the strategic importance of both the origin and destination airport; thus, we use market fixed effects in the simultaneous equation model and the clustered errors at the market level to assess competition effects given the origin and destination airports. However, we do not explicitly model service reliability decisions across multiple markets. We leave these important topics for future research.

Supplemental Material

Supplemental Material, sj-pdf-1-mrj-10.1177_0022243720973943 - Competition and Firm Service Reliability Decisions: A Study of the Airline Industry

Supplemental Material, sj-pdf-1-mrj-10.1177_0022243720973943 for Competition and Firm Service Reliability Decisions: A Study of the Airline Industry by Chen Zhou, Paulo Albuquerque and Rajdeep Grewal in Journal of Marketing Research

Footnotes

Acknowledgments

The authors thank Sriram Venkataraman for helpful comments and feedback, as well as participants of the 2012 Marketing Science conference and seminar participants at Erasmus School of Economics, HEC Paris, Tilburg University, University of Alberta, University of South Carolina, and VU Amsterdam. The article also benefited from the feedback of the JMR review team.

Authors’ Note

This research is based on the doctoral dissertation of the first author.

Guest Editor

Robert Meyer

Associate Editor

S. Sriram

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Paulo Albuquerque

Rajdeep Grewal

Online supplement:

Notes

References

Aguirregabiria

Victor

Mira

Pedro

(2007), “Sequential Estimation of Dynamic Discrete Games,” Econometrica, 75 (1), 1–53.

Air Berlin Group (2008), “Airberlin: Your Airline, Annual Report.”

Alaska Air Group (2012), “Alaska Air Group Sustainability Report,” (accessed December 11, 2020), http://www.flysustainably.com/wp-content/uploads/2018/07/AAG_CSR_2012_Web.pdf.

Allon

Gad

Federgruen

Awi

(2007), “Competition in Service Industries,” Operations Research, 55 (1), 37–55.

Allon

Gad

Federgruen

Awi

Pierson

Margaret

(2011), “How Much Is a Reduction of Your Customers’ Wait Worth? An Empirical Study of the Fast-Food Drive-Thru Industry Based on Structural Estimation Methods,” Manufacturing and Service Operations Management, 13 (4), 489–507.

Anderson

Eugene W.

Mittal

Vikas

(2000), “Strengthening the Satisfaction-Profit Chain,” Journal of Service Research, 3 (2), 107–20.

Anderson

Shannon W.

Baggett

L. Scott

Widener

Sally K.

(2009), “The Impact of Service Operations Failures on Customer Satisfaction: Evidence on How Failures and Their Source Affect What Matters to Customers,” Manufacturing & Service Operations Management, 11 (1), 52–69.

Anderson

Shannon

Pearo

Lisa Klein

Widener

Sally K.

(2008), “Drivers of Service Satisfaction: Linking Customer Satisfaction to the Service Concept and Customer Characteristics,” Journal of Service Research, 10 (4), 365–81.

Bajari

Patrick

Hong

Han

Krainer

John

Nekipelov

Denis

(2005), Estimating Static Models of Strategic Interactions. Ann Arbor: University of Michigan.

10.

Bajari

Patrick

Hong

Han

Krainer

John

Nekipelov

Denis

(2010), “Estimating Static Models of Strategic Interactions,” Journal of Business and Economics Statistics, 28 (4), 469–82.

11.

Bank of America Corporation (2011), “What’s Next for Bank of America, 2011 Annual Report,” (accessed December 11, 2020), http://investor.bankofamerica.com/static-files/443ea392-cd28-4218-93e2-1cab51237aef.

12.

Berry

Leonard L.

(2016), “Revisiting ‘Big Ideas in Services Marketing’ in 30 Years Later,” Journal of Services Marketing, 30 (1), 3–6.

13.

Berry

Steven

Jia

Panle

(2010), “Tracing the Woes: An Empirical Analysis of the Airline Industry,” American Economic Journal: Microeconomics, 2 (3), 1–43.

14.

Brueckner

Jan K.

(2002), “Airport Congestion When Carriers Have Market Power,” American Economic Review, 92 (5), 1357–75.

15.

Cao

Kang Hua

Krier

Betty

Liu

Chia-Mei

McNamara

Brian

Sharpe

Jerrod

(2017), “The Nonlinear Effects of Market Structure on Service Quality: Evidence from the U.S. Airline Industry,” Review of Industrial Organization, 51, 43–73.

16.

Clarke

Michael Dudley Delano

(1998), “Irregular Airline Operations: A Review of the State-of-the-Practice in Airline Operations Control Centers,” Journal of Air Transport Management, 4, 67–76.

17.

Cornwell

Christopher

Trumbull

William N.

(1994), “Estimating the Economic Model of Crime with Panel Data,” Review of Economics and Statistics, 76 (2), 360–66.

18.

Cunningham

Lawrence F.

Young

Clifford E.

Lee

Moonkyu

(2002), “Cross-Cultural Perspectives of Service Quality and Risk in Air Transportation,” Journal of Air Transportation, 7 (1), 3–26.

19.

Cunningham

Lawrence F.

Young

Clifford E.

Lee

Moonkyu

(2004), “Perceptions of Airline Service Quality: Pre and Post 9/11,” Public Works Management & Policy, 9 (1), 10–25.

20.

De Vany

Arthur S.

Saving

Thomas R.

(1983), “The Economics of Quality,” Journal of Political Economics, 91 (6), 979–1000.

21.

Delta Airlines Inc. (2013), “Delta Airlines Inc. Annual Report,” (accessed December 11, 2020), https://www.annualreports.com/HostedData/AnnualReportArchive/d/NYSE_DAL_2013.pdf.

22.

Deshpande

Vinayak

Arikan

Mazhar

(2012), “The Impact of Airline Flight Schedules on Flight Delays,” Manufacturing and Service Operations Management, 14 (3), 423–40.

23.

Durrande-Moreau

Agnes

Usunier

Jean-Claude

(1999), “Time Styles and the Waiting Experience: An Exploratory Study,” Journal of Service Research, 2 (2), 173–86.

24.

Einav

Liran

(2010), “Not All Rivals Look Alike: Estimating an Equilibrium Model of the Release Data Timing Game,” Economic Inquiry, 48 (2), 369–90.

25.

Ellickson

Paul B.

Misra

Sanjog

(2008), “Supermarket Pricing Strategies,” Marketing Science, 27 (5), 811–28.

26.

Ellickson

Paul B.

Misra

Sanjog

(2012), “Enriching Interactions: Incorporating Outcome Data into Static Discrete Games,” Quantitative Marketing and Economics, 10 (1), 1–26.

27.

Elliott

Christopher

(2018), “The Best and Worst Airlines to Fly This Summer,” Forbes (July 21), https://www.forbes.com/sites/christopherelliott/2018/07/21/the-best-and-worst-airlines-to-fly-this-summer/#31509d3b6938.

28.

Ernst

Andreas T.

Jiang

Houyuan

Krishnamoorthy

Mohan

Sier

David

(2004), “Staff Scheduling and Rostering: A Review of Applications, Methods and Models,” European Journal of Operational Research, 153, 3–27.

29.

Ewalt

David M.

(2014), “America’s Best Airlines,” Forbes (April 7), https://www.forbes.com/sites/davidewalt/2014/04/07/americas-best-airlines-2/?sh=57a086b57a18.

30.

Fisher

Marshall L.

Gallino

Santiago

Joseph Jiaqi

(2019), “The Value of Rapid Delivery in Omnichannel Retailing,” Journal of Marketing Research, 56 (5), 732–48.

31.

Forbes

Silke J.

(2008), “The Effect of Air Traffic Delays on Airline Prices,” International Journal of Industrial Organization, 26 (5), 1218–32.

32.

Forbes

Silke J.

Lederman

Mara

(2010), “Does Vertical Integration Affect Firm Performance? Evidence from the Airline Industry,” RAND Journal of Economics, 41 (4), 765–90.

33.

Gijsenberg

Maarten J.

Van Heerde

Harald J.

Verhoef

Peter C.

(2015), “Losses Loom Longer Than Gains: Modeling the Impact of Service Crises on Perceived Service Quality over Time,” Journal of Marketing Research, 52 (5), 642–56.

34.

Greenfield

Daniel

(2014), “Competition and Service Quality: New Evidence from the Airline Industry,” Economic of Transportation, 3 (1), 80–89.

35.

Grieco

Paul L.E.

(2014), “Discrete Games with Flexible Information Structures: An Application to Local Grocery Markets,” RAND Journal of Economics, 45 (2), 303–40.

36.

Hansen

Mark M.

Gillen

David

Djafarian-Tehrani

Reza

(2001), “Aviation Infrastructure Performance and Airline Cost: A Statistical Cost Estimation Approach,” Transportation Research Part E, 37, 1–23.

37.

Harsanyi

John C.

(1973), “Games with Randomly Disturbed Payoffs: A New Rationale for Mixed-Strategy Equilibrium Points,” International Journal of Game Theory, 2 (1), 1–23.

38.

Teck H.

Zheng

Yu-Sheng

(2004), “Setting Customer Expectation in Service Delivery: An Integrated Marketing-Operations Perspective,” Management Science, 50 (4), 479–88.

39.

Irrgang

Jenkins

Ray

(1995), “Airline Irregular Operations,” in Handbook of Airline Economics, Jenkins

Darryl

, ed. New York: McGrawHill Aviation Week Group.

40.

Jhunjhunwala

Pranay

Lee

Jean

Ponce de Leon

Luis

Ramos

Patricio

(2016), “Improving Airlines’ On-Time Performance,” Boston Consulting Group Publications (December 21), https://www.bcg.com/publications/2016/operations-improving-airlines-on-time-performance.

41.

Kalai

Ehud

Kamien

Morton I.

Rubinovitch

Michael

(1992), “Optimal Service Speeds in a Competitive Environment,” Management Science, 38 (8), 1154–63.

42.

Karp

Gregory

(2015), “5 Reasons United Might Finally be Improving Its On-Time Rate,” The Chicago Tribune (November 12), https://www.chicagotribune.com/business/ct-united-airlines-on-time-1113-biz-20151112-story.html.

43.

Katz

Laren L.

Larson

Blaire M.

Larson

Richard C.

(1991), “Prescription for the Waiting-in-Line Blues: Entertain, Enlighten, and Engage,” MIT Sloan Management Review, 32 (2), 44–53.

44.

Levhari

David

Luski

Israel

(1978) “Duopoly Pricing and Waiting Lines,” European Economic Review, 11 (1), 17–35.

45.

Jiang

Liu

Liming

(2012), “Service and Price Competition When Customers Are Naive,” Production and Operations Management, 21 (4), 747–60.

46.

Lode

Lee

Yew Sing

(1994), “Pricing and Delivery-Time Performance in a Competitive Environment,” Management Science, 40 (5), 633–46.

47.

Lohatepanont

Manoj

Barnhart

Cynthia

(2004), “Airline Schedule Planning: Integrated Models and Algorithms for Schedule Design and Fleet Assignment,” Transportation Science, 38 (1), 19–32.

48.

Yina

Musalem

Andres

Olivares

Marcelo

Schikrut

Ariel

(2013), “Measuring the Effects of Queues on Customer Purchases,” Management Science, 59 (8), 1743–63.

49.

Luski

Israel

(1976), “On Partial Equilibrium in a Queueing System with Two Servers,” Review of Economic Studies, 43 (3), 519–25.

50.

Maxon

Terry

(2014), “American Airlines Group to Pay Employee Bonuses if American, US Airways Outdo Operations of Rivals,” Dallas News (January 23), https://www.dallasnews.com/business/airlines/2014/01/23/american-airlines-group-to-pay-employee-bonuses-if-american-us-airways-outdo-operations-of-rivals/.

51.

Mayer

Christopher

Sinai

Todd

(2003), “Network Effects, Congestion Externalities and Air Traffic Delays, or Why All Delays Are Not Evil,” American Economics Review, 93 (4), 1194–215.

52.

Maynard

Micheline

(2013), “The Best and Worst Airlines for Holiday Travel,” Forbes (November 19), https://www.forbes.com/sites/michelinemaynard/2013/11/19/the-best-and-worst-airlines-for-holiday-travel/?sh=4ebabf137de4.

53.

Mazzeo

Michael J.

(2003), “Competition and Service Quality in the U.S. Airline Industry,” Review of Industrial Organization, 22, 275–96.

54.

McCartney

Scott

(2013), “Believe It or Not, Flying Is Improving,” The Wall Street Journal (January 9), https://www.wsj.com/articles/SB10001424127887324081704578231553159491828.

55.

McCartney

Scott

(2014), “A World Where Flights Aren’t Canceled: Inside Delta’s New Strategies to Avoid Stranding Fliers,” The Wall Street Journal (April 2), https://www.wsj.com/articles/SB10001424052702303987004579477412359027986.

56.

McCartney

Scott

(2017), “Which Airlines Pad Their Schedules the Most?” The Wall Street Journal (June 28), https://www.wsj.com/articles/which-airlines-pad-their-schedules-the-most-1498662950#:∼:text=Delta%20and%20United%20add%20more,but%20it%20makes%20a%20difference.

57.

McCartney

Scott

(2018), “The Airlines that Withstood 2018’s Cruel Summer,” The Wall Street Journal (September 5), https://www.wsj.com/articles/the-airlines-that-withstood-2018s-cruel-summer-1536153401.

58.

McKelvey

Richard D.

Palfrey

Thomas R.

(1995), “Quantal Response Equilibria for Normal Form Games,” Games and Economic Behavior, 10 (1), 6–38.

59.

Mittal

Vikas

Anderson

Eugene W.

Sayrak

Akin

Tadikamalla

Pandu

(2005), “Dual Emphasis and the Long-Term Financial Impact of Customer Satisfaction,” Marketing Science, 24 (4), 544–55.

60.

Mittal

Vikas

Kamakura

Wagner A.

Govind

Rahu

(2004), “Geographic Patterns in Customer Service and Satisfaction: An Empirical Investigation,” Journal of Marketing, 68 (3), 48–62.

61.

Mittal

Vikas

Ross

William T.

Baldasare

Patrick M.

(1998), “The Asymmetric Impact of Negative and Positive Attribute-Level Performance on Overall Satisfaction and Repurchase Intentions,” Journal of Marketing, 62 (1), 33–47.

62.

Newey

Whitney K.

McFadden

Daniel

(1994), “Large Sample Estimation and Hypothesis Testing,” in Handbook of Econometrics, Vol. 4, Engle

Robert F.

McFadden

Daniel L.

, eds. Amsterdam: Elsevier, Chapter 36.

63.

Ocampo

Josh

(2019), “The Best and Worst Airlines, Based on Delays and Complaints,” Lifehacker (April 9) https://lifehacker.com/the-best-and-worst-airlines-based-on-delays-and-compla-1833909259.

64.

Orhun

Yesim

(2013), “Spatial Differentiation in the Supermarket Industry: The Role of Common Information,” Quantitative Marketing and Economics, 11, 3–37.

65.

Pai

Vivek

(2010), “On the Factors that Affect Airline Flight Frequency and Aircraft Size,” Journal of Air Transportation, 16 (4), 169–77.

66.

Parasuraman

Zeithaml

Valarie A.

Berry

Leonard L.

(1988), “SERVQUAL: A Multiple-Item Scale for Measuring Consumer Perceptions of Service Quality,” Journal of Retailing, 64 (1), 12–40.

67.

Park

Michael Y.

Vito

Danielle

Allen

Jordan

(2019), “TPG Special Report: The 2019 List of Best and Worst Airlines,”The Points Guy (March 19) https://thepointsguy.com/news/tpg-2019-list-best-and-worst-airlines.

68.

Png

I.P.L.

Reitman

David

(1994), “Service Time Competition,” RAND Journal of Economics, 25 (4), 619–34.

69.

Prince

Jeffrey T.

Simon

Daniel H.

(2015), “Do Incumbents Improve Service Quality in Response to Entry? Evidence from Airlines’ On-Time Performance,” Management Science, 61 (2), 372–90.

70.

Ramdas

Kamalini

Williams

Jonathan

Lipson

Marc

(2013), “Can Financial Markets Inform Operational Improvement Efforts? Evidence from the Airline Industry,” Manufacturing and Service Operations Management, 15 (3), 405–22.

71.

Rupp

Nicholas G.

Holmes

George M.

(2006), “An Investigation into the Determinants of Flight Cancellations,” Economica, 73 (292), 749–83.

72.

Rupp

Nicholas G.

Owens

Douglas H.

Plumly

L. Wayne

(2006), “Does Competition Influence Airline On-Time Performance?” in Advances in Airline Economics, Vol. 1, Lee

Darin

, ed. Bingley, UK: Emerald, 251–72.

73.

Sajtos

Laszlo

Brodie

Roderick J.

Whittome

James

(2010), “Impact of Service Failure: The Perspective Layer of Customer Relationships,” Journal of Service Research, 13 (2), 216–29.

74.

Saporito

Bill

(2014), “Airport Confidential: Inside the Strange World of Airline Cancellations,” Time (March 3), http://content.time.com/time/magazine/article/0,9171,2166055,00.html.

75.

Schlappig

Ben

(2016), “American’s On-Time Performance at LAX Has Improved Greatly,” One Mile at a Time (December 8), https://onemileatatime.com/american-lax-on-time-performance/.

76.

Shang

Weixin

Liu

Liming

(2011), “Promised Delivery Time and Capacity Games in Time-Based Competition,” Management Science, 57 (3), 599–610.

77.

Shirley

Taylor

(1994), “Waiting for Service: The Relationship Between Delays and Evaluations of Service,” Journal of Marketing, 58 (2), 56–69.

78.

Skobelev

Petr

(2011), “Multi-Agent Systems for Real Time Resource Allocation, Scheduling, Optimization and Controlling: Industrial Applications,” in Holonic and Multi-Agent Systems for Manufacturing, Vol. 6867. Berlin: Springer.

79.

Kut C.

(2000), “Price and Time Competition for Service Delivery,” Manufacturing and Service Operations Management, 2 (4), 392–409.

80.

Kut C.

Song

Jing-Sheng

(1998), “Price, Delivery Time Guarantees and Capacity Selection,” European Journal of Operational Research, 111 (1), 28–49.

81.

Sriram

Chintagunta

Pradeep K.

Manchanda

Puneet

(2015), “Service Quality Variability and Termination Behavior,” Management Science, 61 (11), 2739–59.

82.

U.S. Department of Transportation (2014), “Air Travel Consumer Reports for 2014,” (accessed December 11, 2020), https://www.transportation.gov/airconsumer/air-travel-consumer-reports-2014.

83.

Vitorino

Maria Ana

(2012), “Empirical Entry Games with Complementarities: An Application to the Shopping Center Industry,” Journal of Marketing Research, 49 (2), 175–91.

84.

Winston

Clifford

Yan

Jia

(2015), “Open Skies: Estimating Travelers’ Benefits from Free Trade in Airline Services,” American Economic Journal: Economic Policy, 7 (2), 370–414.

85.

Xiong

Jing

Hansen

Mark

(2013), “Modelling Airline Flight Cancellation Decisions,” Transportation Research Part E, 56, 64–80.

86.

Zellner

Arnold

(1962) “An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias,” Journal of the American Statistical Association, 57 (298), 348–68.

87.

Zhu

Ting

Singh

Vishal

(2009), “Spatial Competition with Endogenous Location Choices: An Application to Discount Retailing,” Quantitative Marketing and Economics, 7 (1), 1–35.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.77 MB